Adaptive demodulation based spectral encoded Lempel-Ziv complexity as nonlinear measure for prognostic maintenance of rolling element bearing

As nonlinear measure, additional encoding of original binary Lempel-Ziv complexity (LZC) algorithm has offered promising result in depicting the health condition of rolling element bearing. However, additionally encoded LZC (ELZC) fails to suppress the unwanted random noise in the bearing vibration signal and has always been implemented in analysing time series data while bearing faults are mostly represented by characteristic fault frequencies in spectral domain. As a result, ELZC fails in prognosis of bearing health because it is not only ineffective at identifying the fault at its early stage of occurrence but also fails to track its growth and enable the prediction of its remaining useful life. Aiming to solve aforementioned problems, in this research, a new variant of ELZC namely adaptive demodulation based spectral encoded Lempel-Ziv complexity (ADS-ELZC) is proposed. Firstly, adaptive demodulation of vibration signal is achieved by weighting its squared envelope before calculating the corresponding spectrum. Secondly, instead of time series of the analysed vibration signal, calculated adaptive demodulation based spectrum is encoded to quantify the bearing health for performing its prognostic maintenance operation. Proposed ADS-ELZC not only can detect early incipient bearing fault, but also can track the growth of the incipient fault in a consistent manner along with efficient prediction of remaining useful life even under the small training sample. Verification by two distinct run to failure data infer that proposed ADS-ELZC demonstrate superior performance in comparison to original ELZC, multiscale ELZC (MELZC), Hierarchical ELZC (HELZC) and time series based adaptive demodulated ELZC (AD-ELZC).